| Re: Inverting banded-block matrices. [message #36291] |
Fri, 29 August 2003 17:30 |
the_cacc
Messages: 104
Registered: October 2001
|
Senior Member |
|
|
James Kuyper <kuyper@saicmodis.com> wrote in message news:<3F4E8B4D.B90EDAA8@saicmodis.com>...
> I've got a problem where I have to calculate g = C D^-1 f, where g and f
> are vectors, and C and D are matrices. C has m by m blocks, each of
> which is itself an n by n matrix. It is banded, with k non-zero
> co-diagonals above and below the main diagnal, both at the block level
> and within each block. C[i,j] ge 0. Every statement I've made about C
> also applies to D.
> For the sake of definiteness, m=10, n=1354, k = 3.
>
> This seems like it should be a pretty common type of matrix structure
> for problems involving 2-D grids. I could solve this by explicitly
> inverting a m*n by m*n matrix. However, I would assume that there are
> existing routines somewhere which can take good advantage of the
> sparseness of these matrices to speed up the calculations considerably.
> Could anyone point me at such routines?
There are some sparse routines in IDL - sprsin, sprsax, sprsab,
sprstp. They're "general" rather than specifically for a certain
structure of sparseness, but that should be OK.
Your matrices are pretty big so you can forget about matrix inversion
and use an iterative method instead - conjugate gradient (CG) is good.
Actually, can you even store one of those matrices in memory? (I'm on
a 80 MB machine so I can't!). If you can, then try
IDL> sparse_D = sprsin(D)
(NB. May be rather slow) If you can't, then you'll have to write a
routine to create your matrix in sparse format - hopefully you won't
have to do this - trust me :) Let's leave that alone for the moment.
Rewrite as (1) g = C . h
(2) h = D^-1 . f
The real work is solving (2). You basically have to solve D . h = f,
or sparse_D . h = f, which IDL's linbcg may solve for you.
I found linbcg not so good as the simple CG, which I've implemented.
You'd be welcome to use it if you want - but first check if you can
create the sparse matrix!
Ciao.
|
|
|
|
| Re: Inverting banded-block matrices. [message #36304 is a reply to message #36291] |
Fri, 29 August 2003 08:32  |
Liam E. Gumley
Messages: 378
Registered: January 2000
|
Senior Member |
|
|
"James Kuyper" <kuyper@saicmodis.com> wrote in message
news:3F4E8B4D.B90EDAA8@saicmodis.com...
> I've got a problem where I have to calculate g = C D^-1 f, where g and f
> are vectors, and C and D are matrices. C has m by m blocks, each of
> which is itself an n by n matrix. It is banded, with k non-zero
> co-diagonals above and below the main diagnal, both at the block level
> and within each block. C[i,j] ge 0. Every statement I've made about C
> also applies to D.
> For the sake of definiteness, m=10, n=1354, k = 3.
>
> This seems like it should be a pretty common type of matrix structure
> for problems involving 2-D grids. I could solve this by explicitly
> inverting a m*n by m*n matrix. However, I would assume that there are
> existing routines somewhere which can take good advantage of the
> sparseness of these matrices to speed up the calculations considerably.
> Could anyone point me at such routines?
I believe LAPACK routines were incorporated in IDL 5.6, however I have not
tried them.
Cheers,
Liam.
Practical IDL Programming
http://www.gumley.com/
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
